Hewlett Packard Enterprise Co. and Nvidia Corp. today announced a new set of offerings aimed at simplifying the deployment and scaling up of artificial intelligence models in government agencies, regulated industries and large enterprises.

The joint portfolio, unveiled at Nvidia’s GTC conference in Washington, D.C., principally address data fragmentation, infrastructure sprawl and compliance needs.

The centerpiece is the second generation of HPE’s Private Cloud AI, a smaller version of the turnkey platform aimed at fast and secure deployments. The goal is to help organizations “move together as one and be able to cross across environments and focus on the business outcomes, not on the infrastructure,” said Robin Braun, vice president of AI business development for hybrid cloud at HPE.

Smaller footprint

The updated system integrates HPE’s ProLiant DL380a Gen12 servers and Nvidia’s new RTX PRO 6000 Blackwell Server Edition graphics processing units. Braun said it delivers three times better price-performance compared to the prior generation, based on the MLPerf benchmark.

HPE is also adding an air-gapped management option for organizations with strict data isolation requirements. An air-gapped server operates on an isolated network with no external internet or network connections, making it harder for unauthorized users or malware to access sensitive data.

One of the first practical deployments of the infrastructure is in Vail, Colorado, The city is piloting HPE’s Agentic Smart City, an agentic AI framework that leverages specialized, autonomous AI agents to manage citywide systems in real time.

The project leverages AI orchestration software from Kamiwaza Corp., image recognition from ProHawk Technology Group Inc. and geospatial analysis from Blackshark.ai GmbH.

HPE said Vail aims to streamline public services and improve accessibility, housing, safety and digital interaction. “There are around 4,800 residents on a yearly basis,” Braun said. “However, they get about 2.8 million visitors a year. That flux in the environment allows us to start with a great pilot and then scale it out as we work with other cities.”

Unified data

To address the challenge of fragmented data in AI training, HPE also introduced enhancements to its unified data layer that enabled data governance using a agentic AI and unstructured data storage. The system combines structured, semistructured and unstructured data under a single federated namespace.

It integrates with HPE’s Alletra Storage MP X10000 and now includes support for remote direct memory access over object storage, a feature developed in collaboration with Nvidia. The new capability “reduces latency by up to 80%, and CPU utilization by up to 99%,” said Gokul Sathiacama, vice president of unstructured data. The storage platform can also operate in air-gapped environments and is now available with for use in on-premises private clouds.

To meet the demands of training larger AI models, HPE introduced the ProLiant XD685 server, supporting eight Nvidia Blackwell Ultra B300 GPUs and using direct liquid cooling. Also new is the Nvidia GB300 NVL72 by HPE, a rack-scale system that ties together 72 Blackwell GPUs into a single logical GPU using Nvidia NVLink. It’s designed for models with more than 1 trillion parameters.

HPE said it continues to expand GPU options across its broader server portfolio, including configurations for the RTX A1000 and other entry-level accelerators aimed at smaller organizations and lighter AI workloads.

The two companies also revealed a new partnership with the University of Utah, which HPE said is aimed at delivering a sovereign AI infrastructure to support statewide research efforts.

Photo: HPE